Le 19/09/2014 15:36, Daniel Stenberg a écrit :
> On Thu, 18 Sep 2014, KALLEL Mohamed wrote:
>
>> I want to add the support of sending http messages over UDP to libcurl.
>
> What are you talking to in the other end? I know there have been talks
> about doing HTTP over UDP but I'm curious about the implementation
> situation among servers.
In fact I do not have a big idea about http protocol neither about http
over udp cases. But I see this request in the standard protocol TR-069
cwmp (Amendment 5) and in this standard they require to send HTTP
messages over UDP and they are not asking to expect answer from the server.
Ref of TR-069:
http://www.broadband-forum.org/technical/download/TR-069_Amendment-5.pdf
(Annex M)
>
>> Please let me know if my patch is OK? and If you are going to merge
>> it to the mainline? and in what version will be delivered?
>
> Let's take one step at a time and see where we end up!
>
> First, you merged two rather independent changes into a single patch
> and that's not a good idea. Especially since I strongly dislike the
> second part - the DISABLE_RECEIVE thing. You need to provide a
> convincing argument why we would want that.
You are right for the 2 patches issue. I will send you 2 separate
patches for that. For the DISABLE_RECEIVE, I used it because for my case
I really do not care if the server answer to my UDP message or not, So
that's wy I want to disable the receive part in the process.
>
> Then, your patch doesn't come with any documentation or test cases
> even though it introdces a new option (and we really can't have that
> unless you have a really good excuse to avoid it). I take it you have
> run your own tests against something and verified that this actually
> works exactly like it does when using TCP? What happens when you send
> (really) large requests?
For the documentation I can send you documentation with the next patch.
please provide to me a template of your documentation. For the tests, I
made test for my case and it works properly but for the libcurl test
cases, I really do not have idea neither exeprience with that
>
> The name of the option. CURLOPT_UDP_CONNECT sounds like an oxymoron to
> me. UDP doesn't "connect" at all in the first place. Your code
> explicitly use this to send HTTP over UDP so shouldn't the option name
> say that? Like for example CURLOPT_HTTP_OVER_UDP ? Or possibly if we
> would allow it for other protocols too, CURLOPT_USE_UDP ?
> CURLOPT_FORCE_UDP?
CURLOPT_FORCE_UDP sound good. It was the first name I choosed ;-)

Regards
MOHAMED Kallel
EasyCwmp maintainer
www.easycwmp.org

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2014-09-23